Recent developments in Alzheimer's disease (AD) and aging research suggest that reducing future cases may be feasible by preventing AD in high-risk individuals by initiating treatments prior to neurodegeneration. Identifying such individuals requires characterizing the earliest stages of dementia when individuals are still asymptomatic and early disease effects are weak. To detect subtle (faint yet persistent) effects, large sample sizes are typically needed, but existing studies are not sufficiently large to provide high statistical power. Motivated by this critical need, this project will develop the statistical theory and algorithms needed to pool multi-cohort Alzheimer's disease datasets to facilitate large scale statistical analysis in samples of cognitively unimpaired individuals who are still asymptomatic. The development will be driven by (a) new shape analysis methods to characterize brain anatomy at the individual level and (b) statistical machine learning algorithms to enable seamless pooling of data from different studies/cohorts. Specific Aim 1: Develop novel shape analysis methods which enable testing advanced hypotheses in a way that is largely invariant to study-wise biases in Alzheimer's disease datasets collected at two different research sites (DELCODE, WRAP/WADRC) using compact descriptors of individual level brain anatomy (neuroanatomical signatures). Specific Aim 2: Derive new statistical theory and algorithms, based on classical statistical constructs and deep learning, for harmonizing shape features and other clinical/cognitive features focused on early Alzheimer's disease across the two sites. Specific Aim 3: Perform association and prediction analyses on the pooled datasets to evaluate novel scientific hypotheses related to early stages of Alzheimer's disease. We will analyze cerebrospinal fluid (CSF) biomarkers, shape features and longitudinal cognition (slopes/intercepts) and finally conduct analyses to better understand heterogeneity and sub-groups within the pooled cohort of cognitively unimpaired adults. Significance: This project will (i) lead to the development of efficient shape representation frameworks that will allow testing advanced localized hypotheses on brain regions of interest without requiring sub-field segmentation. The impact of these shape analysis methods will support spatially localized brain image analysis in Alzheimer's disease and in aging studies more generally. (ii) Produce harmonization algorithms and statistical theory to enable large scale data pooling in Alzheimer's disease as well as a broad range of other aging and neuroscience studies.